[SPARK-40059] Row indexes can overshadow user-created data

• Type: Improvement
• Status: Open
• Priority: Major
• Resolution: Unresolved
• Affects Version/s: 3.4.0
• Component/s: SQL
• Assignee: Unassigned
• Reporter: Ala Luszczak
• Created: 12/Aug/22 15:11
• Updated: 12/Aug/22 15:11

Description

https://github.com/apache/spark/pull/37228 introduces ability to compute row indexes, which users can access through `_metadata.row_index` column. Internally this is achieved with the help of an extra column `_tmp_metadata_row_index`. When present in the schema sent to parquet reader, the reader populates it with row indexes, and the values are later placed in the `_metadata` struct.

While relatively unlikely, it's still possible, that a user might want to include column `_tmp_metadata_row_index` in their data. In such scenario, the column will be populated with row indexes, rather than data read from the file.

For repro, search `FileMetadataStructRowIndexSuite.scala` for this Jira ticket number.

We could introduce some kind of countermeasure to handle this scenario.

Issue Links

• is related to: SPARK-37980 Extend METADATA column to support row indices for file based data sources